Semantics
Lexical Ambiguity
When one word form has multiple meanings — homonymy and polysemy
Lexical ambiguity occurs when a single word form maps to multiple distinct senses. "Bank" can mean a financial institution or the side of a river — homonymy (unrelated senses, accidental). "Mouth" can refer to a body part or the opening of a river — polysemy (related senses, motivated by metaphor or metonymy). The distinction was sharpened by Stephen Ullmann (1957) and George Lakoff. English has thousands of ambiguous words: "set" has 430+ senses in the OED, the most of any English word. Lexical ambiguity is resolved by context (semantic priming), syntactic frame, and pragmatic inference. It is a constant challenge for natural language processing — word sense disambiguation (WSD) is a benchmark NLP task. Senseval (1998) and SemEval (since 2007) are standard evaluation campaigns.
- Two main typesHomonymy (unrelated) vs polysemy (related)
- Most polysemous English word"set" — 430+ OED senses
- ResolutionContext, syntactic frame, world knowledge
- NLP taskWord Sense Disambiguation (WSD)
- Standard benchmarkSemEval campaigns
- Historical theoristStephen Ullmann (1957)
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
Why lexical ambiguity matters
- NLP and search. Disambiguation drives accurate retrieval.
- Translation. Wrong sense yields wrong translation; context-aware MT critical.
- Lexicography. Sense distinctions are the dictionary's main product.
- Psycholinguistics. Reveals the architecture of mental lexicon.
- Reading instruction. Multi-sense words are vocabulary teaching pivots.
- Wordplay and humor. Puns exploit ambiguity systematically.
- Legal interpretation. Statutory ambiguity is a recurring courtroom issue.
Common misconceptions
- Ambiguity is rare. Most common words are highly polysemous.
- Context fully resolves all cases. Some ambiguities persist (intentional puns, headlines).
- Homonymy and polysemy are easy to distinguish. The boundary is fuzzy and historically driven.
- Ambiguity slows comprehension noticeably. Resolution is automatic and fast (<200ms).
- Ambiguity is a defect. It's a feature: efficient lexical packaging.
- One sense is always primary. Frequency biases shift across registers and time.
Frequently asked questions
What's the difference between homonymy and polysemy?
Homonymy: same form, unrelated meanings. "Bank" (river) and "bank" (financial) come from different historical sources — Old English "banca" vs Italian "banca." Polysemy: same form, related meanings — typically through metaphor or metonymy. "Head" of a body, "head" of a department, "head" of a beer — all related to a topmost-or-leading concept. Dictionaries reflect this: homonyms get separate entries; polysemes get one entry with numbered senses.
How do listeners resolve it?
Context-driven. Hearing "I deposited money at the bank" — only the financial sense is possible. Resolution happens within ~200ms of word onset (Swinney 1979). Initially, all senses activate briefly (multiple-access hypothesis), then context selects. Frequency biases: "bug" usually means insect, not microphone — listeners preferentially activate dominant sense first.
What are syntactic disambiguators?
Syntactic frame constrains sense. "I saw her duck" — verb reading ("saw her duck under the bridge") vs noun reading ("saw her pet duck"). The verb reading requires "her" as object; noun requires it as possessive. Verbs select different argument structures: "drink" (consume liquid) takes a beverage; "drink" (consume alcohol heavily, intransitive) takes nothing.
Which English words have the most senses?
According to the Oxford English Dictionary: "set" (~430), "run" (~645 senses in the 2011 OED revision — surpassing set), "go" (~370), "take" (~340), "stand" (~330). Common, short, Anglo-Saxon-origin verbs dominate the top of the polysemy ranking. They acquire senses over centuries through metaphor and grammaticalization.
What's word sense disambiguation in NLP?
The task of assigning the correct sense to an ambiguous word in context. Approaches: (1) supervised — train classifiers on sense-tagged corpora (SemCor, OntoNotes). (2) knowledge-based — use WordNet, Lesk algorithm, gloss overlap. (3) unsupervised clustering. (4) modern — contextual embeddings (BERT, RoBERTa) inherently encode sense in context. WSD accuracy on standard benchmarks now exceeds 80%.
How does ambiguity differ from vagueness?
Ambiguity has discrete senses ("bat" = animal vs sports equipment). Vagueness has fuzzy boundaries ("tall" — how tall?). Test: zeugma (mixing senses sounds odd if ambiguous, fine if vague). "He bought a bat and a glove" (sports senses fine), but "He bought a bat and ate a sandwich" forces sense selection. Vagueness doesn't pattern this way.
What about cross-linguistic mismatches?
Polysemy patterns vary. English "head" extends to "head of state, head of beer, head of cabbage." Spanish "cabeza" doesn't fully cover these — "cabeza de Estado" works, "cabeza de cerveza" doesn't. Cross-linguistic polysemy networks (Wälchli & Cysouw 2012) reveal universal vs language-specific extensions. Major source of translation difficulty.